Proposal: Scenario-based reserve sensitivity analysis

[son1403]

I would like to discuss scenario-based reserve sensitivity analysis.

Teams commonly calculate a base reserve and then rerun the process manually for alternative development or tail assumptions. The comparison is important, but the repeated work makes it harder to keep the scenarios consistent.

A helper could take an estimator or pipeline and named scenarios, refit each scenario, and show the movement in ultimate and IBNR against the base. It may be better as an extension of the existing model-selection tools rather than a separate workflow.

The closed prototype is #1250. I have opened this issue first to settle the right home and the smallest useful scope.

[henrydingliu]

It may be better as an extension of the existing model-selection tools rather than a separate workflow.

what about as a new property of the votingchainladder emsemble?

[son1403]

It may be better as an extension of the existing model-selection tools rather than a separate workflow.

what about as a new property of the votingchainladder emsemble?

Yes, that seems like a better fit. I had been thinking about making the effect of alternative development or tail selections visible, but a standalone scenario runner is probably broader than necessary. Would you also expect it to compare alternative weightings, or should the first version only describe the fitted blend? I can put together an API proposal and example once that boundary is clear.

[henrydingliu]

based on your description, i was visualizing something like this. each row would be a single scenario. the columns would be ultimate, ibnr, or anything other financial metric we want to compare. the actual weighted estimate would be another row (see WeightedEnsemble_L2 in the linked table above).